Skip to content
MyAI Guide
Newsletter
Hugging Face models and Vercel tools dominate AI launches | Daily AI roundup cover
Daily RoundupIndustryVibe BuilderNon TechnicalDeveloper

Hugging Face models and Vercel tools dominate AI launches

By Harsh Desai
Share

TL;DR

On 20 May multiple multimodal models hit Hugging Face and Replicate while Vercel expanded its AI SDK and WordPress plugin; Google pushed agentic ads and creative tools. Builders gain new sandboxes, local memory bridges, and unified model access without custom integrations.

What shipped

On 20 May several multimodal models climbed Hugging Face charts and Replicate added Gemini 3.5 Flash. Vercel released SDK updates and a WordPress connector while Google Marketing Live highlighted agentic ad formats and creative studios. These releases give teams direct paths to test image-to-video pipelines and permission-aware agents in production.

Hugging Face trending

Four models gained rapid traction on the Hub this week. Two image-text-to-text checkpoints from CohereLabs and HauhauCS joined an image-to-video model from Efficient-Large-Model and a video-text model from NemoStation. Each ships ready for fine-tuning and inference, letting builders prototype multimodal agents without managing separate runtimes.

  • •CohereLabs checkpoint CohereLabs released command-a-plus-05-2026-w4a4, an image-text-to-text model trending on Hugging Face. Vibe builders can fine-tune it for custom captioning tasks and deploy via the Hub API in minutes.
  • •Efficient-Large-Model video model Efficient-Large-Model launched SANA-WM_bidirectional, an image-to-video model now climbing the Hub charts. Teams can generate short clips from stills and iterate on marketing assets without separate video pipelines.
  • •NemoStation video model NemoStation released Marlin-2B, a video-text-to-text model trending on Hugging Face. Developers can connect it to existing agents for scene description and cut manual review time.
  • •HauhauCS uncensored model HauhauCS published Gemma4-26B-A4B-Uncensored-HauhauCS-Balanced, an image-text-to-text model gaining attention on the Hub. SMB owners can test open-ended visual queries while staying within their own data boundaries.

Vendor launches

Vercel added AI SDK tools to Chat SDK and released a WordPress plugin that routes any AI plugin to 40 providers through one key. Google Marketing Live introduced Gemini-powered ad formats, Asset Studio multimodal edits, and Ask Advisor for turning campaign data into actions. Grok Build 0.1 also joined the Vercel AI Gateway for agentic coding tasks.

  • •Chat SDK tools Vercel added a chat/ai subpath to Chat SDK so one createChatTools call connects read and write actions to agents. Vibe builders can gate write tools behind approval and ship scoped agents faster.
  • •Google I/O recap Google listed Gemini Omni, Google Antigravity, and Universal Cart among 100 announcements at I/O 2026. Builders can review the full set to pick multimodal features for upcoming projects.
  • •AI Search ads Google launched new Gemini-based ad formats in Search and expanded the Direct Offers pilot. Marketers can test intent-driven creatives that respond to live shopping queries.
  • •WordPress AI plugin Vercel released an AI Gateway plugin that gives any WordPress site access to hundreds of models from 40 providers via one key. Non-technical site owners can activate AI features without writing provider code.
  • •Grok Build on Gateway xAI added Grok Build 0.1 to Vercel AI Gateway as a beta coding model for agentic tasks. Developers can call it through the unified API and track usage across retries.
  • •Ask Advisor agent Google introduced Ask Advisor, a cross-product agent that keeps a running thread of insights. Teams can convert campaign data directly into next-step recommendations.
  • •Marketing Live collection Google Marketing Live 2026 presented AI solutions for growth across Google and YouTube. SMB owners can adopt Demand Gen and creator tools to lift video performance.
  • •Universal Commerce updates Google added UCP features and AI tools for more intuitive shopping flows. Retailers can let agents handle product discovery and reduce manual catalog work.
  • •Asset Studio upgrades Google brought multimodal capabilities to Asset Studio for creating and testing ad assets. Creative teams can generate variations from text and image prompts in one workspace.
  • •YouTube Demand Gen Google expanded Demand Gen with Gemini AI and Maps reach for turning viewer intent into conversions. Marketers can run campaigns that blend short video with location signals.
  • •AI Studio expansions Google AI Studio added native Android vibe coding, Workspace integrations, and a mobile app. Vibe builders can move prototypes from browser to device without extra setup.
  • •Running Guide agent Google DeepMind released Running Guide, an audio navigation agent for BLV athletes. Developers can study its real-time obstacle detection as a reference for mobile agent design.

Replicate new models

Gemini 3.5 Flash on Replicate: Google released gemini-3.5-flash on Replicate for frontier reasoning across agents and coding. Builders can run long-context tasks through the HTTP API and compare latency against prior versions.

Product Hunt picks

Seven new tools appeared on Product Hunt for sandboxed agents, local memory, screenshot compression, and security workflows. Runtime and Glia target team coding and browser-to-IDE handoff while Owlish and Hiro focus on support and compliance agents.

  • •Runtime sandboxes Runtime launched sandboxed coding agents that any team member can run. Vibe builders can test agent code without local setup and share sessions across projects.
  • •Gemini Omni tool Google introduced Gemini Omni for creating content from any input starting with video. Creators can generate follow-on assets directly from source clips.
  • •Glia memory bridge Glia released a local-first memory bridge between browser chats and IDEs. Developers can keep context across tools without sending data to external servers.
  • •GhostSnap compressor GhostSnap added auto-compressed multi-screenshot capture for AI workflows. Users can feed clean image sets into vision models with less manual editing.
  • •Supercut access control Supercut introduced permission-aware access to recordings and metadata for agents. Teams can audit agent activity while limiting scope to approved data.
  • •Owlish support agent Owlish released AI agents trained on company docs to cut support volume. SMB owners can route common questions to the agent and free staff for complex tickets.
  • •Hiro security agent Hiro launched an agentic security team that runs from first commit to SOC 2. Developers can embed compliance checks into every pull request without separate tooling.

What this means for you

For Vibe Builders: New Hugging Face checkpoints from CohereLabs and Efficient-Large-Model let you fine-tune captioning and image-to-video models through the Hub API in minutes, and Runtime on Product Hunt runs sandboxed coding agents any teammate can share with no local setup. Google AI Studio added native Android vibe coding, so you can move a prototype from browser to phone. Start with one image-to-video model like SANA-WM to test whether it replaces a manual asset-creation step.

For Non-techies: The new Vercel AI Gateway plugin for WordPress gives your site access to hundreds of models through one key, with no provider code to write. On the marketing side, Google's Ask Advisor turns campaign data into next-step recommendations, and Owlish answers common support questions from your own docs. Try Owlish first if support tickets are eating your team's time.

For Developers: Gemini 3.5 Flash landed on Replicate for long-context reasoning, and Grok Build 0.1 joined Vercel AI Gateway as a beta coding model you can call through one unified API with usage tracking across retries. Vercel's Chat SDK now ships createChatTools so you can gate agent write actions behind approval. Hiro's agentic security flow embeds SOC 2 and compliance checks into every pull request, worth a look before you wire agents into production accounts.

What to watch next

Track adoption numbers for the new CohereLabs and Efficient-Large-Model checkpoints on Hugging Face. Watch Vercel AI Gateway usage metrics after the WordPress plugin release. Note any early SOC 2 or audit reports from teams using Hiro or Supercut.

Harsh’s take

Most launches this week repackage existing model access behind simpler connectors rather than delivering new capabilities. Google ads and creative tools dominate volume while smaller teams gain practical sandboxes and memory bridges that actually change daily workflows. The real signal sits in how quickly developers adopt local-first options like Glia instead of adding another cloud hop.

Builders should pick one new connector, such as the Vercel WordPress plugin or Runtime sandbox, and run a two-day internal test against current manual processes. Measure setup time and output quality before committing to any wider rollout.

by Harsh Desai

Sources

Hugging Face trending

Vendor launches

Replicate new models

Product Hunt picks

More AI news

Everything AI. One email.
Every Monday.

New tools. Model launches. Plugins. Repos. Tactics. The moves the sharpest builders are making right now, before everyone else.

No spam. Unsubscribe anytime.